Zyphra sells open frontier AI models, the infrastructure to run them, and a productivity agent. The site positions the company around the idea of open superintelligence, where companies choose, customise and control the intelligence they use.
The product line has four parts. Zyphra Cloud offers datacenter capacity, accelerators, networking and storage across AMD, Google TPU, AWS Trainium and Nvidia hardware, and the site asks visitors to contact them for capacity. The Zyphra Optimization Engine includes model and hardware specific kernels, algorithms and routing for inference. Open Models and Inference lets customers use Zyphra and other open-source frontier models in the cloud or in a sovereign environment. The Customization Studio, which the site marks as in development, is meant to turn general open models into custom ones through fine-tuning, reinforcement learning, and an integrated harness and memory.
MAIA is a multiplayer agent for knowledge workers. According to the site, it remembers shared context, takes action across tools and helps teams turn knowledge into outcomes. It is integrated with a sovereign stack and custom intelligence, or available as an application. MAIA is in beta, and visitors can join the beta list.
The site also publishes research on inference speed, attention methods and mixture-of-experts training, with source code for some of the work. It lists claimed results such as 3.6 times faster time to first token and 60 percent lower cost per session, though these figures are presented on the homepage without detail on how they were measured.
The company is aimed at businesses. Its solution pages cover enterprises, startups and frontier hyperscale customers. The site does not state pricing for its products. Cloud capacity is quoted on request, and MAIA is in beta.
Zyphra is based in San Francisco, and the site gives an office address there. The footer names Zyphra Technologies Inc.
